As someone who was once a bench scientist, I'd love to think I was immune to these kinds of pitfalls. After all, we control things carefully in the lab, we can

As someone who was once a bench scientist, I'd love to think I was immune to these kinds of pitfalls. After all, we control things carefully in the lab, we can control many more of the inputs to an experiment, test a hypothesis by different avenues, have truly independent controls... But this is a very large literature, that in effect by sheer volume of work was doing many of those things. This is quite humbling. That said, I do think that medical science, though massively flawed (see previous discussion on clinical trials no longer revealing positive results https://news.ycombinator.com/item?id=11209778 https://news.ycombinator.com/item?id=11209778) there is a way to make it so we can trust science. By: 1.) as previously said given a hypothesis, it must be confirmed by 2 independent approaches (if this is true, then this other thing should be true, is it?) 2.) publishing more negative results, removing the incentive for the file-drawer effect mentioned by @jobu 3.) require much larger sample sizes for studies to be published, and take effect size properly into account when judging whether the outcome is valid. One interesting way to do this: http://www.stat.columbia.edu/~gelman/research/published/retropower_final.pdf http://www.stat.columbia.edu/~gelman/research/published/retr...